File size: 9,535 Bytes
23cdeed 66ad25b 0a62245 66ad25b 0a62245 66ad25b 0a62245 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 | # -*- coding: utf-8 -*-
"""
pluto/tools.py — Corpus access tools (spec §3).
Implements list_docs, search, get_chunk, get_figure, get_table, log, finish
over a local corpus/ directory.
"""
from __future__ import annotations
import json
import os
import re
from pathlib import Path
from typing import Any
from pluto.tracer import Tracer
class CorpusTools:
"""File-backed implementation of the spec's external tool interface."""
def __init__(self, corpus_dir: str, output_dir: str = "./output", tracer: Tracer | None = None, doc_index=None) -> None:
self.corpus_dir = Path(corpus_dir).resolve()
self.output_dir = Path(output_dir).resolve()
self.output_dir.mkdir(parents=True, exist_ok=True)
self.tracer = tracer
self.doc_index = doc_index # DocIndex instance (if available)
self._doc_cache: dict[str, str] = {}
self._chunk_cache: dict[str, list[str]] = {} # doc_id -> list of chunks
# ── list_docs ──────────────────────────────────────────────────────────
def list_docs(self) -> list[dict[str, str]]:
"""Return metadata for every document in the corpus."""
docs = []
for f in sorted(self.corpus_dir.iterdir()):
if f.suffix in (".md", ".txt", ".pdf"):
docs.append({
"doc_id": f.stem,
"filename": f.name,
"size_bytes": str(f.stat().st_size),
})
if self.tracer:
self.tracer.log("list_docs", {"count": len(docs)})
return docs
# ── search ─────────────────────────────────────────────────────────────
def search(self, query: str, filters: dict | None = None) -> list[dict[str, Any]]:
"""
Semantic search across all documents using NVIDIA NIM reranker.
Falls back to keyword scoring if reranker is unavailable.
"""
if self.tracer:
self.tracer.record_search(query)
self.tracer.log("search", {"query": query})
allowed_doc_ids = None
if filters and filters.get("doc_ids"):
allowed_doc_ids = {
str(doc_id).strip()
for doc_id in filters.get("doc_ids", [])
if str(doc_id).strip()
}
# Collect all candidate passages
candidates = []
for f in sorted(self.corpus_dir.iterdir()):
if f.suffix not in (".md", ".txt"):
continue
if allowed_doc_ids is not None and f.stem not in allowed_doc_ids:
continue
content = self._read_doc(f.stem)
# Use first 500 chars of doc as the candidate for doc-level scoring
candidates.append({
"doc_id": f.stem,
"snippet": content[:500],
"full": content,
})
if not candidates:
return []
# Try NIM reranker first
try:
from pluto.dispatcher import rerank
passages = [c["snippet"] for c in candidates]
scores = rerank(query, passages)
for c, s in zip(candidates, scores):
c["score"] = s
except Exception:
# Fallback: keyword scoring
keywords = query.lower().split()
for c in candidates:
c["score"] = sum(c["full"].lower().count(kw) for kw in keywords)
candidates.sort(key=lambda x: x["score"], reverse=True)
return [
{"doc_id": c["doc_id"], "score": c["score"], "snippet": c["snippet"][:300]}
for c in candidates[:20]
]
# ── get_chunk ──────────────────────────────────────────────────────────
def get_chunk(self, doc_id: str, chunk_id: str) -> str:
"""Return the source text of a specific chunk for extraction."""
chunks = self.get_all_chunks(doc_id)
if self.tracer:
self.tracer.record_doc_opened(doc_id)
self.tracer.log("get_chunk", {"doc_id": doc_id, "chunk_id": chunk_id})
try:
idx = int(chunk_id.lstrip("C"))
except ValueError:
return ""
if 0 <= idx < len(chunks):
return strip_non_extractable_context(chunks[idx])
return ""
def get_all_chunks(self, doc_id: str) -> list[str]:
"""Return all chunks for a document (cached after first split)."""
# Check DocIndex first (pre-indexed at upload)
if self.doc_index and self.doc_index.has_doc(doc_id):
return self.doc_index.get_chunks(doc_id)
# Fallback: split on-the-fly + cache
if doc_id not in self._chunk_cache:
content = self._read_doc(doc_id)
self._chunk_cache[doc_id] = self._split_into_chunks(content)
return self._chunk_cache[doc_id]
# ── get_figure ─────────────────────────────────────────────────────────
def get_figure(self, doc_id: str, figure_id: str) -> str | None:
"""Return path to a figure image if it exists."""
for ext in (".png", ".jpg", ".jpeg", ".svg"):
p = self.corpus_dir / f"{doc_id}_{figure_id}{ext}"
if p.exists():
return str(p)
return None
# ── get_table ──────────────────────────────────────────────────────────
def get_table(self, doc_id: str, table_id: str) -> str:
"""Return table text extracted from the document."""
content = self._read_doc(doc_id)
tables = re.findall(
r"(\|.+\|(?:\n\|.+\|)+)",
content,
re.MULTILINE,
)
idx = int(table_id.replace("T", "")) if table_id.startswith("T") else 0
if 0 <= idx < len(tables):
return tables[idx]
return ""
# ── log ────────────────────────────────────────────────────────────────
def log(self, event: str, payload: dict[str, Any]) -> None:
"""Append event to the trace log."""
if self.tracer:
self.tracer.log(event, payload)
# ── finish ─────────────────────────────────────────────────────────────
def finish(self, final_json: dict) -> Path:
"""Write final JSON output to disk."""
out_path = self.output_dir / "final_output.json"
out_path.write_text(json.dumps(final_json, indent=2, ensure_ascii=False), encoding="utf-8")
if self.tracer:
self.tracer.log("finish", {"output_path": str(out_path)})
return out_path
# ── Internal helpers ───────────────────────────────────────────────────
def _read_doc(self, doc_id: str) -> str:
if doc_id in self._doc_cache:
return self._doc_cache[doc_id]
for ext in (".md", ".txt"):
p = self.corpus_dir / f"{doc_id}{ext}"
if p.exists():
text = p.read_text(encoding="utf-8")
self._doc_cache[doc_id] = text
return text
return ""
def _split_into_chunks(self, content: str, max_chunk: int = 1500) -> list[str]:
"""Split document into chunks by headings or paragraph groups."""
# Split on markdown headings first
sections = re.split(r"\n(?=#+\s)", content)
chunks: list[str] = []
for section in sections:
section = section.strip()
if not section:
continue
if len(section) <= max_chunk:
chunks.append(section)
else:
# Further split on double newlines
paras = section.split("\n\n")
current = ""
for para in paras:
if len(current) + len(para) + 2 > max_chunk and current:
chunks.append(current.strip())
current = para
else:
current += "\n\n" + para if current else para
if current.strip():
chunks.append(current.strip())
return chunks if chunks else [content]
def strip_non_extractable_context(chunk_text: str) -> str:
"""Remove metadata prefixes that must not be treated as document evidence."""
text = str(chunk_text or "").lstrip()
patterns = (
r"^\[Document context:[^\]]*\]\s*",
r"^\[Context\s*\|[^\]]*\]\s*",
)
changed = True
while changed:
changed = False
for pattern in patterns:
cleaned = re.sub(pattern, "", text, flags=re.IGNORECASE | re.DOTALL)
if cleaned != text:
text = cleaned.lstrip()
changed = True
return text
|